An Improved Assessment of Poisson’s Distribution and Applications
Himpriya Kumari and Ashok Pal
Department of Mathematics, Chandigarh University, Gharuan, Mohali - 140413
Email address:24msm10007@[Link],[Link]@[Link]
ABSTRACT
his paper provides a comprehensive analysis of the Poisson Distribution, including its historical
T
background and past developments. We examineitspropertiesandexploreitsreal-lifeapplications,
discussing key concepts such as mean, variance, and moments. Additionally, we delve into the
Poissonprocess,itsproperties,anditsapplications.ThePoissonDistributionisparticularlyusefulin
analyzingscenariosliketrafficflow,predictingfaultsinelectriccables,andestimatingtheoccurrence
ofrandomaccidents.Itisespeciallyrelevantincaseswhereeventsarerarewithinalargesampleor
observationspace.Furthermore,thereisacloserelationshipbetweenthePoissondistributionandthe
binomial distribution, where the former can be derived as an approximation of the latter when the
numberoftrialsislargeandtheprobabilityofsuccessissmall.Thepaperalsoaddressesextensionsof
thePoissonmodel,suchastheZero-InflatedPoissonandNegativeBinomialdistributions,whichhelp
manage limitations like overdispersion and excess zeros, making the model adaptable to more
complexdatascenarios.Overall,thePoissondistributionremainsafundamentaltoolforthestatistical
modelling of count data, supporting analysesinbothscientificandindustrialcontexts.Wewillalso
explore the limitationsandassumptionsinherenttothisdistributionthroughoutthepaper,especially
focusingonitsappropriatenessincontextswhereeventsarerareacrosslargeobservationspaces,and
we will discuss the relationship between the Poisson and binomial distributions in depth.
EYWORDS – Probability distribution, binomial distribution, discrete distribution, Poisson
K
distribution
1 INTRODUCTION
he Poisson distribution is an important concept in probability and statistics, commonly used to
T
represent the frequency of events occurring over a specific period or area. Developed by Siméon
Denis Poisson in the early 19th century, it was initially designed to analyze rare events, such as
unlikelyaccidentsornaturaloccurrenceswithindefinedtimeframes.Overtime,ithasevolvedintoa
vitaltoolforunderstandingandpredictingeventfrequenciesacrossvariousfields,includingphysics,
biology, telecommunications, finance, and public health. [5]
neofthekeyfeaturesofthePoissondistributionisitsapplicabilitytosituationswhereeventsoccur
O
randomly,independently,andataconsistentaveragerate.Thismakesitparticularlyrelevantfordata
related to infrequent or unpredictable events, such as the number of meteor sightings in a night,
incoming calls at a call centre, or system failures during a specified time. It is mathematically
expressedthroughasingleparameter,λ(lambda),whichrepresentsboththemeanandvarianceofthe
distribution. [4]
espiteitsstraightforwardnature,thePoissondistributionisbasedonseveralassumptions,including
D
theindependenceofeventsandaconstantrateofoccurrence.However,real-worlddataoftendeviates
from these assumptions, leadingtothedevelopmentofextensionssuchastheZero-InflatedPoisson
(ZIP) and the Negative Binomial distribution. These adaptations address challenges like
overdispersion or the presence of excesszeros,ensuringthatthePoissonmodelremainsrelevantin
c ontemporarystatisticalanalysis,especiallyinscenarioswherethereissignificantvariabilityinevent
frequency. [6]
I n conclusion, the Poisson distribution is an essential tool for statistical modelling, bridging
theoretical principles with practical applications. Its wide range of uses, along with advanced
extensions, highlights its importance in tackling modern data analysis challenges in areas where
predicting events and modelling counts are crucial.
2 LITERATURE SURVEY
hePoissondistribution,introducedbytheFrenchmathematicianSiméonDenisPoissonin1837,isa
T
statistical concept that describes the probability of a certain number of events occurring within a
specifiedtimeframeorarea.TheoriginsofthePoissondistributionarelinkedtoearlierdevelopments
inmathematicsandthegrowinginterestinprobabilitytheoryduringthe18thand19thcenturies.This
distribution emerged fromPoisson’sinvestigationsintoprobabilitytheoryasitpertainedtocriminal
and civil legal cases. His findings, published in "Recherches sur la Probabilité des Jugements en
Matière Criminelle et en Matière Civile," represent one of the early attempts to probabilistically
model the frequency of rare events. ThePoissondistributiongainedpracticalinterestandempirical
support through the work of Ladislaus Bortkiewicz, a Russian statistician of Polish descent. Inhis
1898 publication, "Das Gesetz der Kleinen Zahlen" ("The Law of Small Numbers"), Bortkiewicz
analyzeddataregardingthenumberofsoldierswhodiedfromhorsekicksinthePrussianarmyovera
specified period. He discovered that the Poisson distribution effectively modelled these rare,
seemingly random events. Despite the infrequency of horse kick fatalities, Bortkiewicz found that
theyoccurredatarelativelyconsistentrateovertheyears,aligningwellwiththeassumptionsofthe
Poissonmodel.Hisresearchhelpedvalidatethedistributionasareliabletoolforanalyzingrareevents
and introduced it to a broader statistical audience. [3]
3 POISSON DISTRIBUTION
3.1Definition-AvariablexissaidtohavethePoissondistributionifittakesthevalues0,1,2,…to
2
−𝑚 𝑚 −𝑚 𝑚 −𝑚
with probabilities𝑒 , 1! 𝑒 , 2! 𝑒 ,...., 𝑡𝑜∞ respectively. (See figure 1.1) [1]
Note: Poisson Distribution has only one parameter m.
.
Figure 1.1.
robability Mass Function (PMF) - A discrete irregular variable is said to have a Poisson
P
Distribution with boundary m>0, if for x=0, 1, 2, …. at that point the likelihood mass capacity is
given by:
𝑥 −𝑚
𝑚 𝑒
P(x) = 𝑥!
here e is Euler’s number (e=2.178…) x is the number of occurrencesandx!isthefactorialofx.
w
[2][6]
A graphical representation of PMF is shown in Figure number 1.2.. [5]
Figure 1.2.
3.2 Mode of Poisson Distribution
The mode of Poisson Distribution is the value of the variate x which gives the greatest probability.
herefore, the probability of (x-1) successes ≤ probability of x successes ≥ probability of (x+1)
T
successes
−𝑚 (𝑥−1) −𝑚 𝑥 −𝑚 (𝑥+1)
𝑒 𝑚 𝑒 𝑚 𝑒 𝑚
i.e., (𝑥−
1)!
≤ 𝑥!
≥ (𝑥+
1)!
−1
𝑚 1 𝑚
or, (𝑥−1)! ≤ 𝑥{(𝑥−
1)!}!
≥ (−1)𝑥{ (𝑥−
1) !}
−1
𝑚
1 𝑚
or, 1
≤ 𝑥
≥ (𝑥+
1)𝑥
These give𝑚 − 1 ≤ 𝑥 ≤ 𝑚
I fmisnotanintegerthenthemodeisanintegralpartofmandifmisaninteger,thentherearetwo
modes one at (m-1) and the other at m and the probabilities in both these cases are equal. [1]
3.3 Parameters of Poisson Distribution
3.3.1 Moments About Origin -
∞ 𝑥
' −𝑚 𝑚
[Link] First moment about the origin -µ1 = ∑ 𝑒 𝑥!
𝑥 = 𝑚
𝑥=
0
∞ 𝑥
' −𝑚 𝑚 2 2
[Link] Second moment about the origin -µ2 = ∑ 𝑒 𝑥!
𝑥 = 𝑚 + 𝑚
𝑥=
0
∞ 𝑥
' −𝑚 𝑚 2 3 2
[Link] Third moment about the origin -µ3 = ∑ 𝑒 𝑥!
𝑥 = 𝑚 + 3𝑚
+ 𝑚
𝑥=
0
∞ 𝑥
' −𝑚 𝑚 2 4 3 2
[Link] Fourth moment about the origin -µ4 = ∑ 𝑒 𝑥!
𝑥 = 𝑚 + 6𝑚
+ 7𝑚
+ 𝑚
𝑥=
0
3.3.2 Moments about Mean
[Link] First moment about mean -µ1 = 0
' '2
[Link] Second moment about mean -µ2 = µ2 − µ
= 𝑚
Variance=µ2 = 𝑚andStandard Deviation= 𝑚
' ' ' '3
[Link] Third moment about mean -µ3 = µ3 − 3µ
2µ1 + 2µ
1 = 𝑚
' ' ' ' '2 '4 2
[Link] Fourth moment about the mean-µ4 = µ4 − 4µ
3µ1 + 6µ
2µ1 − 3µ
1 = 3𝑚
+ 𝑚
Note:The graphical representation of Poisson Distribution. [2]
Figure 1.3.
𝑡
𝑡𝑋 (𝑚(𝑒 −1))
3.3 Moment Generating Function - 𝑀𝑋𝑡 = 𝐸( 𝑒 ) = 𝑒 , where m is the mean (and
variance) of the Poisson Distribution.
3.4 Pearson’s coefficient -
2
µ3 2
𝑚
β1 = 3 = 3
µ2 𝑚
1
or, β1 =
𝑚
1
Kurtosis =β1 =
𝑚
µ4 2
3𝑚 +𝑚
β2 = 2 = 2
µ2 𝑚
1
or, β1 = 3 +
𝑚
1
γ1 = β1 =
𝑚
1
Skewness =γ1 =
𝑚
1
γ2 = β2 − 3 =
𝑚
ote:Apoisondistributionispositivelyskewed.However,thedegreeofskewnessdecreaseswithan
N
increase in the value of λ. See in figure 1.3.. Here λ = m (parameter). [2]
Figure 1.4.
4 Methodologies Utilizing Poisson Distribution
he Poisson distribution is commonly used to modeltheprobabilityofaspecificnumberofevents
T
occurring within a fixed interval of time or space. This is applicable when the events happen
independently of one another and at a constant average rate. Below are some methodologies and
approaches that are frequently associated with the Poisson distribution:
(These methodologies are referred from [2], [4] and [6]).
4.1 Parameter Estimation
● Maximum Likelihood Estimation (MLE): For a Poisson distribution with parameter m
𝑛
1
(average rate of events), MLE estimates m as the sample mean,𝑚= 𝑛
∑ 𝑥𝑖.
𝑖=1
● B ayesianEstimation:Assumesapriordistributionform(oftenGammadistributionasitis
conjugatetoPoisson),thencombinesitwithobserveddatatoformaposteriordistributionfor
m.
4.2 Goodness of Fit Testing
● C hi-square Test: Used to determine if observed frequencies match expected frequencies
under the Poisson model.
● Likelihood Ratio Test (LRT): Compares the likelihood of data under a Poisson model
against an alternative hypothesis model.
4.3 Modelling Overdispersion
● W hen data show more variability than a Poisson model predicts, Negative Binomial
DistributionorQuasi-Poisson modelsare often used.
● Zero-inflated Poisson (ZIP)andHurdle Modelshelp model data with excess zeros.
4.4 Regression Analysis with Poisson Models
● P oissonRegression:Forcountdatainfluencedbypredictorvariables,thelogoftheexpected
count is modelled as a linear combination of the predictors.
● Generalized Linear Models (GLMs): Poisson GLMs use a log-link function, enabling
regression analysis for count data.
4.5 Time Series and Spatial Poisson Models
P
● oisson Processes:Continuous-time processes usedto model counts over time or space.
● Spatial Poisson Models: Used for modelling spatial data where the location of events is
random and follows a Poisson process.
4.6 Simulation Techniques
● M onte Carlo Simulation: Generate synthetic data to understand and validate Poisson
distribution models.
● Bootstrapping: For generating confidence intervals or distributions of Poisson-based
estimates.
hesemethodologiesofferflexibilityinmodellingdifferentscenariosusingthePoissondistribution.
T
They ensure that the approach can adapt to data exhibiting extra variability, dependencies on
predictors, or specific spatial-temporal patterns. The choice of each technique depends on the data
structure, underlying assumptions, and the specific requirements of the research or application.
5 POISSON DISTRIBUTION AS A LIMITING FORM OF BINOMIAL DISTRIBUTION
Statement: Poisson Distribution is the limiting form of the binomial distribution
𝑛
(𝑞 + 𝑝) 𝑤ℎ𝑒𝑛𝑝(𝑜 𝑛𝑑𝑛→ ∞, suchthatnp(ornq)isafinitequantitysaymi.e.,np=
𝑟𝑞) → 0𝑎
m. (See Figure 1.5.) [1]
Figure 1.5.
6Benchmark Problems Employing Poisson Distribution
(These examples have been referred from [1])
6.1.What is the probability of throwing an ace with a die exactly 10 times in 1200 trials?
Solution. The probability p of throwing an ace with a die = (⅙)
Also number of trials = n = 1200
Therefore, if the Poisson distribution is used then the mean m = np = 1200 * (⅙) = 200.
−𝑚 10
𝑒 𝑚
Therefore, the required probability of throwing an ace is exactly 10 times = (10)!
, where m = 200.
−200 10
Therefore, probability is = [𝑒 * (200) ]/(10)!
.2.Amanufacturerofscrewsknowsthat5%ofhisproductisdefective.Ifhesellstheminboxesof
6
100andguaranteesthatnomorethan10screwswillbedefective,whatistheapproximateprobability
that a box will fail to meet the guaranteed quality?
Solution. Here p = (5/100); n = 100 so m = np = 100* (5/100) = 5.
lso, we know that according to Poisson distribution (here p is very small), the probability of x
A
screws being defective is given by
−𝑚 𝑥 −5 𝑥
𝑒 𝑚 𝑒 5
(𝑥) !
i.e., (𝑥) !
10 −5 𝑥
𝑒 5
Therefore, the probability of ten or less than ten screws to be defective = ∑ (𝑥) !
.
𝑥=0
10 −5 𝑥
𝑒 5
Therefore, the required probability of the defective screws to be more than ten = 1 - ∑ (𝑥) !
.
𝑥=0
7 APPLICATIONS
hePoissondistributionisastatisticalmodelfordescribingrandomandindependenteventswithina
T
fixed timeframe or area. Here are key applications across various fields:
These applications are referred from [4], [5] and [6].
7.1 Telecommunications
. Call Centers: The number of calls received at a call centre per hour often follows a Poisson
1
distribution, which helps optimize staffing.
.NetworkTraffic:ThearrivalofdatapacketsatrouterscanbemodelledasaPoissonprocess,aiding
2
in congestion management.
7.2 Healthcare and Epidemiology
. Disease Incidence: New cases of a rare disease within a specific timeframe typically follow a
1
Poisson distribution, assisting in predicting outbreaks.
. Emergency Room Demand: Hospitals use the modeltoestimatedailypatientarrivals,helpingin
2
resource allocation.
7.3 Environmental Science
.NaturalEvents:RareeventslikeearthquakescanbemodelledusingPoissondistributiontopredict
1
occurrence likelihoods.
2. Wildlife Conservation: Tracking sightings of rare species helps in conservation assessments.
7.4 Manufacturing and Quality Control
. Defect Counts: The number of defects inabatchcanbeestimatedusingPoissondistributionfor
1
quality control.
2. Machine Failures: Equipment breakdowns over time can also follow this distribution.
7.5 Finance and Insurance
1. Claim Counts: Insurance companies model claim filings using Poisson distributions to assess risks.
. Bank Transactions: The number of daily transactions can follow a Poisson process for cash
2
management.
7.6 Transportation and Traffic Flow
. Accidents: The occurrence of traffic incidents often aligns with a Poisson distribution, helping
1
improve safety measures.
.BusArrivals:ThenumberofarrivalsatbusstopsmodelledasaPoissonprocessaidsinoptimizing
2
schedules.
7.7 Crime and Public Safety
. Crime Incidents: Police use Poisson distribution to model crime occurrences, aiding in resource
1
allocation.
.FireCalls:Thenumberoffireincidentsisfrequentlymodelledusingthisdistributionforeffective
2
resource management.
7.8 Sports and Gaming
.GoalsinSoccerMatches:ThenumberofgoalsscoredcanfollowaPoissondistribution,assisting
1
in predicting outcomes.
. Player Performance: Specific events in games, like three-pointers, can also be modelled using
2
Poisson.
7.9 Biology and Genetics
1. Mutations: Genetic mutations during cell division can be modelled using the Poisson distribution.
2. Cell Biology: Tracking cell division or death in experiments often uses this statistical approach.
8 CONCLUSION AND FUTURE PROSPECTS
I nconclusion,thePoissondistributionisavaluabletoolforpredictingrareorrandomeventsinmany
fields, such as healthcare, environmental science, manufacturing, and public safety. It works well
becauseitcaptureshowofteneventshappenoversetperiods.Thismakesitidealforeventsthatoccur
independently and at a steady average rate.
y using Poisson models, organizations and researchers can allocate resources better, streamline
B
processes, and make smart decisions. This helps them meet demand, improve safety, and boost
efficiency. The Poisson distribution is essential foranalyzingtheoccurrenceofrandomeventsover
time or space. It provides a clear way to assess how likely certain counts are when these events
happen independently and at a regular rate.
he future of the Poisson distribution appears promising, thanks to its growing integration with
T
advanced analytics, machine learning, and real-time data applications. As technology and data
generation continue to accelerate, the Poisson model is expected to evolve significantly, enabling
complex decision-making, predictive analysis, and automation. .
I ts simplicity and flexibility make it especially helpful for counting data. This includes tracking
diseasecases,managingserviceresources,orstudyingrareeventsinresearch.Ultimately,thePoisson
istributionconnectstheoreticalprobabilitywithreal-lifeapplications.Itgivesasolidfoundationfor
d
data-driven planning and forecasting.
REFERENCES
[1] S. Gupta and V. Kapoor, Fundamentals of Mathematical Statistics. Sultan Chand & Sons,
2020.
[2] “HELM Workbook 37 Discrete Probability Distributions.” [Link] 2008.
[Link] (accessed: 2008).
[3] “UNIT 10 POISSON DISTRIBUTION.” eGyanKosh.
[Link] (accessed: Nov. 05, 2024).
[4] “An Extension of the Poisson Distribution: Features and Application for Medical Data
Modeling.” MDPI. Apr. 28, 2023.[Link]
[5] “A Short Review on Poisson Distribution with Some Applications and ..” [Link].
[Link] (accessed: Nov. 06, 2024).
[6] “The Properties and Application of Poisson Distribution - IOPscience.” [Link].
[Link] (accessed: Nov. 06, 2024).